Auctions and Market Design - MIT OpenCourseWare€¦ · hands of billion-dollar poker at once. •...
Transcript of Auctions and Market Design - MIT OpenCourseWare€¦ · hands of billion-dollar poker at once. •...
Auctions and Market Design
Alessandro Bonatti
MIT Sloan
Today’s Agenda
1) The Theory of Engineering Games
2) Designing Markets
3) Designing Auctions
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 2
Game Theory Approach
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 3
Fixed Game Equilibrium Outcome(s)
“solver”
Mechanism Design Approach
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 4
Desired Outcome
Game that delivers it
as…
“designer”
… Nash equilibrium? Dominant Strategy? …
Connected Applications
1. NYC Public Schools 2. Medical Residents 3. Spectrum auctions worldwide
• More applications later
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 5
NYC Schools pre-2004 • The goal: to place each student in the best school for him/her. • The problem: you don’t know students’ preferences.
• Students submit a list of five schools. Schools select admits. • The higher performing students get multiple offers. • Nearly half the students (not a random sample) got no offers • Thousands of unlucky teenagers wound up waiting through the
summer only to be sent to schools they had not listed at all. • Those schools were worse in all dimensions — including
student achievement, graduation rate and college admissions — than the schools the students had asked to attend.
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 6
The Cost of Complexity • The system encouraged safe, not ambitious, choices. • Sought-after schools accepted only the applicants
who had made them their first choice • Students who aimed high and listed several such
schools but were rejected by the first could blow their chances all the way down the list.
• Education Department’s high school directory advised students to “determine what your competition is for a seat in this program"
• Now go figure out the Nash Equilibrium!!
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 7
NYC Schools Today
• Students list their 12 favorite schools, in order of preference • The algorithm allows students to “propose” to their favorite
school, which accepts or rejects the proposal. • In the case of rejection, the algorithm looks to make a match
with a student’s second choice school, and so on. • Students and schools connect only tentatively until the very
end of the process. • Unmatched/year from 31,000 to about 3,000. • 50% top choice school and 85% top 3 choices • Why?
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 8
Costs of a Decentralized Market • Hospitals benefit from filling positions early • Applicants benefit from delaying acceptance of positions. Offers were being made for positions up to 2 years in advance. Efforts were made to delay the start of the process. Very short deadlines for responses by applicants Opportunities for dissatisfaction … for both…
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 9
Centralized Admissions
• Each applicant ranks “acceptable” hospitals
• Each hospital ranks “acceptable” candidates
• Both can be (essentially) as long as you want.
• Rankings are submitted to a centralized system
• NRMP runs “the algorithm.”
• Gale and Shapley (1962): “College Admissions and the
Stability of Marriage”
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 10
Deferred-Acceptance Algorithm 1. Each applicant “proposes” to his/her favorite hospital 2. The hospital rejects all applicants except its favorite
one, but does not accept him/her just yet. 3. Each rejected applicant proposes to his/her second-
choice hospital. 4. Each hospital again rejects all but its favorite applicant
(possibly throwing the old favorite back in the pool). 5. The “proposals” continue until the process stops
(i.e., no current favorite is replaced any more).
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 11
“List them in order of true preference”
• The resulting match is stable: no student-hospital pair that did not form would prefer to now break their current matches and get together.
Why?
• It is a dominant strategy for an applicant to submit his / her true ranking.
Why?
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 12
Several Difficulties
• Hospitals manipulation – “In the limit as the market becomes infinitely large, the proportion
of employers who might profit from preference or capacity manipulation goes to zero (Kojima and Pathak, 2009)
• Couples
– This problem does not go away with size
• …
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 13
Hospital Incentives • It is not necessarily optimal for hospitals to rank applicants truthfully. • Example: three students, (A B C); two hospitals (1 and 2). • Student A prefers hospital 1 to 2 • Student B prefers hospital 2 to 1 • Student C prefers hospital 2 to 1 • Hospital 1 prefers student B to A to C • Hospital 2 prefers student A to B to C
• If everyone reports truthfully, the students get their top choices in the
first round. So the outcome is (A,1) and (B,2). • If hospital 2 instead reports (ACB), then B and C apply to 2;
hospital 2 keeps C and throws B in the pool. B applies to 1, which keeps him/her. Finally, A applies to 2, which keeps him/her.
• So the outcome is (A,2), and (B,1) which hospital 2 prefers. Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 14
Judicial Law Clerks (failed)
Current guidelines: • Support a transparent recruitment process. • Consider coordinating hiring activities court-wide. • Use video conferencing when feasible. • Do not require an applicant to accept an offer immediately
without reasonable time to weigh it against other viable offers. • Consider visiting law schools with a minority student
population. How well do you think this works?
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 15
Spectrum Auctions (FCC’s Solution)
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2014 16
FCC auction • Game theory, long an intellectual pastime, came into
its own as a business tool.
• 1993: Congress authorizes FCC to auction spectrum for personal communications services.
• Goals: simplicity, efficiency, revenue, and diversity.
• Thousands of licenses varying in geographic coverage and amount of spectrum.
• Bidders: long-distance, local and cellular telephone companies.
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2014 17
FCC auction • Auction format: simultaneous ascending auction
– multiple licenses are open for bidding as long as there is some bidding on any one.
• Bidding for the PCS licenses is like playing a dozen hands of billion-dollar poker at once.
• Closing the gaps for clever bidders: details matter!
• Most important: activity rule (33-66-100)
• Secondary: bid increments.
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2014 18
FCC auction • Why ascending (~English)?
– Winner’s curse
• Why simultaneous?
– Licenses are interdependent
– Resale market
– Gaming and holding-up
• Still: jump-bidding and signaling.
• Total revenues > $60 billion (1994, ~$100B today)
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2014 19
What’s the Connection?
• All licenses are open • Bidders have preferences over licenses • “Licenses” only care about $$$ • Standing high-bidder = current favorite • Next round, could be replaced… • … but has the option to “relocate” • … or to reclaim its spot (up to the valuation)
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 20
Other Applications
• West Point Cadets
• Kidney Exchanges
• Course allocation
• Financial markets
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 21
High-Frequency Trading
• Clearing demand and supply for financial assets. • Periodic auctions? • “Instantaneous” order execution?
• Instantaneous: transparent, simple, dominant
strategies, no “transaction risk,” but …
• Perverse incentives: “dumb money” in the deal • Overinvestment….
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 22
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 23
Light Speed?
• In 2010, Spread Networks completed construction of a new high-speed fiber optic cable
• Connecting financial markets in New York and Chicago. • Spread Networks’ cable was nearly a straight line. • Construction costs were $300 million. • The result of this investment? • Round-trip communication time between New York and
Chicago was reduced . . . ….from 16 milliseconds to 13 milliseconds.
• Relativity bound is > 7 milliseconds
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 24
Market Design: General Principles
• Transparency
• Strategy-proofness (dominance)
• Stability
• Simplicity
• Design elements complement each other
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 25
Next Week
• Games with Private information – Signaling – Communication – Poker
• R&D and project selection in Pharma
Prof. Alessandro Bonatti MIT Sloan 15.025 Spring 2015 26
MIT OpenCourseWarehttp://ocw.mit.edu
15.025 Game Theory for Strategic AdvantageSpring 2015
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.